Getting started

git clone https://github.com/alonso59/fundus_suitable.git  # clone
cd fundus_suitable
conda install pytorch torchvision cudatoolkit -c pytorch
pip install -r requirements.txt  # install

Embedded repository YOLO V5 for retina detection FORK

cd src/
git clone https://github.com/alonso59/retina_det.git

Prepare config file YAML

Classification

model_name: 'inceptionv3' # inceptionv3, resnet18, vgg13, swin_custom, swin_tiny, nat_mini, nat_custom

Segmentation

model_name: 'unet' #unet, unet_imagenet, swin_unet, swin_unet_custom

Operation modes

Training, Evaluation & Predict

cd ..
python src/main.py --stage class --mode train --config configs/classifier.yaml           
                           pre           eval          configs/segmenter.yaml                           
                           det           predict  

Dataset classification

├── datasets/
    classification/
        ├── train/
            ├── 0/
            ├── 1/
        ├── val/
            ├── 0/  
            ├── 1/
        ├── test/
            ├── 0/
            ├── 1/

Dataset segmentation

├── datasets/
    segmentation/
        ├── train/
            ├── Images/
            ├── Masks/
        ├── val/
            ├── Images/  
            ├── Masks/

Dataset detection

    python datasets/detection/perpare_files.py

├── datasets/
    detection/
        ├── images/
        ├── labels/
        ├── train.txt
        ├── val.txt

Implementation

    python src/main.py --stage impl --source dataset/images/
                                             dataset/images/input.jpg

Citation

@misc{pinedo,
      title={Fundus image quality assesment}, 
      author={Pinedo-Diaz},
      year={2022},
}